Modeling the time dependence of the association between human papillomavirus infection and cervical cancer precursor lesions

The authors studied the time-dependent association between human papillomavirus (HPV) infection and squamous intraepithelial lesions (SIL) among women enrolled in a cohort study in Brazil (1993-2002), using repeated Papanicolaou cytologic examination and HPV testing by polymerase chain reaction. Through simulation with conceivable alternative cohort designs, they investigated different regression modeling approaches using time-varying covariates, time-varying hazard ratio functions, and repeated events to assess the effect of delay in lesion detection. Associations between HPV and early SIL were of high magnitude. The age-adjusted hazard ratios for the association between HPV at enrollment and low-grade SIL decreased gradually with time until 72 months for both oncogenic types of HPV (hazard ratio = 3.96, 95% confidence interval (CI): 2.5, 6.4) and nononcogenic types (hazard ratio = 2.37, 95% CI: 1.3, 4.3). The hazard ratio for incident high-grade SIL remained constant, ranging from 7.15 (95% CI: 2.0, 25.1) at 12 months to 6.26 (95% CI: 2.7, 14.5) at 72 months for oncogenic types of HPV. With oncogenic HPV as the time-dependent predictor variable, the hazard ratios for incident SIL and high-grade SIL events were 14.2 (95% CI: 8.7, 23.1) and 32.7 (95% CI: 8.4, 127.3), respectively. Investigators may underestimate the prognostic value of HPV detection using designs that rely on HPV ascertainment at a single time point. The waning in hazard ratios should be considered in the implementation of HPV testing-based screening programs.